A modeling framework for infrastructure planning of Workflow-as-a-Service environments

Abstract

Given the characteristics of dynamic provisioning and the illusion of unlimited resources, “the cloud” is becoming a popular alternative for running scientific workflows. In a cloud system for processing workflow applications, the performance of the system is heavily influenced by two factors: the scheduling strategy and failure of components. Failures in a cloud system can simultaneously affect several users and depreciate the number of available computing resources. A bad scheduling strategy can increase the expected makespan and the idle time of physical machines. In this work, we propose a modeling framework, and a set of formal models and methods for supporting the infrastructure planning of Workflow-as-a-Service clouds. This modeling framework supports the tasks of: i) planning the deployment of workflow applications in computational clouds in order to maximize performance and reliability metrics; ii) planning the redundancy arrangements in the cloud infrastructure in order to reduce the acquisition cost while satisfying availability requirements; iii) identifying availability bottlenecks and enabling the prioritization of critical components for improvement. We conducted three case studies in order to illustrate and validate the proposed modeling framework. The first case study employs a comprehensive hierarchical availability model using RBD and DRBD models and applied sensitivity analysis methods in order to find the most influential parameters. The second case study extends the previous one by modeling a cloud infrastructure as an instance of the redundancy allocation problem (RAP). To minimize the acquisition cost while maximizing the availability of the system, we proposed the combined use of a local-search algorithm and the bisection method. In the last case study, we optimize the scheduling of scientific cloud workflows. This case study comprises the use of a metaheuristic algorithm coupled with a performability model that provides the fitnesses of the explored solutions. The experimental results obtained in all case studies have proven the framework effectiveness on aiding planning infrastructures tasks, allowing cloud providers to maximize resource utilization, reduce operational costs, and ensure SLA requirements.